The cruellest divide in tomorrow’s workplace: people who direct agents and people replaced by them
I came across the news that ClickUp had laid people off the other day. At first, I did not think much of it.
We have become a little numb to layoffs at tech companies. Today one company says it is improving efficiency; tomorrow another says it is refocusing on its core business. Translated into plain language, it often means: money is tight, so we have to shrink.
But one detail about this round at ClickUp made me pause.
It laid off 22% of its employees.
Then its CEO, Zeb Evans, came out and said this was not about saving money.
His explanation was that the company wanted to embrace AI more aggressively and turn ClickUp into a “100x org.”
Even more striking, he said the company would introduce a million-dollar compensation band. Anyone who could use AI to create outsized value would no longer be constrained by traditional pay structures.
My first reaction was: man, that sounds so Silicon Valley.
Lay people off, say it is not about saving money, then say that those who remain may earn a million dollars.
A few years ago, I probably would have dismissed this as the usual capital-markets story: listen, then move on.
This time, though, I hesitated.
Because there was another detail in the TechCrunch article.
ClickUp had recently introduced around 3,000 internal AI agents. Employees would no longer just do the work themselves; they would direct the agents and check their output.
That was the part I found genuinely interesting.
To be frank, layoffs themselves are nothing new.
What is new is that companies are beginning to ask a question more seriously: should a person’s value still be measured by how much work they personally complete?
Think about it. We used to measure productivity in units of people.
How many calls does a salesperson make in a day?
How many pieces of copy does a marketing operator write?
How many tickets does a customer-service representative handle?
How many features does a programmer deliver in a sprint?
Of course, companies bought software: CRMs, Notion, ClickUp and other collaboration systems. But most of the time, these tools were extensions of people.
People were still the main characters.
Software just helped them move faster, remember things more reliably and avoid some coordination headaches.
But agents are changing the unit of work.
You are no longer just one person.
You are more like a small command center.
You break a task down and hand different pieces to different agents. They research, draft, run processes, organize meetings, make spreadsheets and investigate problems. Then you come back to inspect the work: what is wrong, what is usable and what needs to be thrown out and rebuilt.
That may sound abstract, right?
Let me use my own work as an example.
When I write an article now, I often do not sit in front of a blank document and force the words out. I might ask Deep Research to gather a company’s recent moves, ask ChatGPT to find similar cases, ask Claude Code or Codex to organize local materials, or have different models challenge one another’s conclusions.
But I am still the one who decides whether the article works.
I have to know whether the topic is interesting.
I have to know which facts are mere hooks and which details actually matter.
I have to know whether a passage sounds like something a person would say.
More difficult still, I have to know when to stop.
This is genuinely counterintuitive.
Many people think knowing how to use AI means knowing how to write prompts.
Prompts do matter, of course, but they are only the entrance.
The harder questions are whether you can break a task down into work a machine can handle; whether you can tell if the machine’s answer is really an answer; whether you can turn a successful process into something reusable instead of starting over next time; and whether you can stop a machine when it confidently starts making things up.
That is the new productivity.
In the past, a strong employee was someone who could do a lot themselves.
In the future, a strong employee may be someone who can do the work and get a team of tireless execution units to do it alongside them.
Damn.
That is a big difference.
So when I saw the number of 3,000 internal agents at ClickUp, my first thought was not “another company using AI as an excuse to lay people off.”
Of course, that possibility is always there.
I completely understand why people resent this kind of narrative.
An ordinary employee reading the news probably will not feel excited.
They may just think: great, the company has found another new excuse.
It used to be “cost reduction and efficiency.” Now it is “agent-native.”
It used to be “organizational transformation.” Now it is “100x org.”
Either way, I am still the one who gets the axe.
That feeling is real.
You are not the founder.
You are not an investor.
You are not one of the people shouting on X every day that the future is already here.
Maybe you are just someone who works hard at a company: meetings during the day, documents to catch up on at night, messages to answer over the weekend. You are already tired enough. Now you are also expected to learn how to direct AI, automate your own role and prove you are not the person who can be replaced.
I really understand that exhaustion.
But for that very reason, I do not think we can understand this only as “the company is laying people off again.”
It is more like a sharp-edged reminder.
Organizations are recalculating what “useful” means.
Before, getting the work done was enough to make you useful.
Soon, just getting the work done may not be enough.
The company will keep asking: have you made this easier to do next time?
Have you turned a manual process into a partly automated one?
Have you turned your judgment into a workflow other people can reuse?
Have you let an agent take over the repetitive, trivial and uncreative parts?
Honestly, this change is a little cruel.
But it is also a little fair.
In many organizations, the people who create the most value are not always the ones most visible.
Some people are busy all day and appear to produce a lot, but they are really just moving the same bricks over and over.
Others are quieter. They can straighten out a chaotic process, reduce rework for the team, help new colleagues get up to speed faster and turn something repeated ten times a month into a one-time configuration.
In the past, that kind of ability was often hard to quantify.
The boss knew you were good but could not say exactly why.
The team depended on you, but the performance review had nowhere to put it.
Now that agents are entering the picture, this ability may become more visible.
Once you can break work down clearly, spell out the criteria for judgment and define the inputs and outputs, a machine really can take over part of it.
You are not just doing one task.
You are building a system that can do tasks repeatedly.
That is why ClickUp’s idea of “gamifying not token costs, but the value created and time saved” caught my attention.
Many companies measure AI adoption by looking at how much people use it.
Whoever consumes the most tokens or makes the most calls looks the most enthusiastic.
But that can go badly wrong.
It is like saying someone is a great communicator because they send 200 WeChat messages a day.
Or that someone is AI-native because they leave ChatGPT open all day.
What we should look at is what they made faster.
Which errors did they reduce?
Which process no longer depends on someone logging in at 2 a.m. to put out a fire?
What task that used to involve three people arguing back and forth can one person and a few agents now get to a 70-point first version?
That 70-point version matters.
Some people will say agents are unstable, hallucinate and still fail at complex tasks.
Yes.
Absolutely.
I am still figuring them out myself, and they often leave me speechless.
Sometimes I ask one to research something. It gives me an answer that sounds incredibly convincing, so I click through and find the page says nothing of the sort.
Sometimes I ask one to edit an article, and it sands away the little bit of human life I most wanted to keep, turning it into a very dignified year-end report.
I am left speechless.
But a lot of work in organizations does not need to be perfect on the first pass anyway. It needs a 70-point draft.
Gather the materials first.
Fill in the spreadsheet first.
List the candidate options first.
Organize the meeting notes first.
Then a person can apply judgment, make trade-offs, exercise taste and take responsibility.
That is what I think is really worth discussing about ClickUp.
It is not “AI replacing people.”
That phrase is too crude.
More precisely, people are moving up the execution chain.
You are no longer just the person tightening the screws by hand.
You have to know which screws need tightening, which do not, how tight is good enough, and whether a problem came from the design or the execution.
That sounds like management.
But it is not management in the traditional sense.
Traditional management manages people.
In the future, ordinary roles will also have to manage machines.
Sometimes I think this may create a strange new class divide.
Not managers versus workers.
People who can orchestrate, and people who can only be orchestrated.
That sounds a little harsh, but the more I think about it, the harder it is to avoid.
People who can orchestrate can extend their judgment beyond themselves, even if they do not have a senior title. They can use Codex to write scripts, Claude Code to change a project, ChatGPT to research, and Excel or Feishu tables to turn a process into a system. They do not have to be programmers. They just need to know how to get a workforce of digital laborers moving toward a goal.
People who can only be orchestrated are in a more dangerous position.
Not because they do not work hard.
Maybe they work extremely hard.
But if all that effort stops at “I did it by hand,” it will be squeezed by more and more semi-automated workflows.
Picture this.
Employee A handles 20 customer issues a day.
Employee B can initially handle only 20 too, but spends two weeks turning common questions into templates, cleaning up the knowledge base, setting boundaries for the agent’s answers and writing rules for when unusual cases go to a person.
Two weeks later, B is still one person.
But there is a small system behind them.
They may cover 80 issues a day and tell the boss which issues come up most often, which answers prompt the most follow-up questions, and which product defects should be sent back to engineering.
Who do you think the company will choose?
It is cold, but realistic.
This does not mean everyone should turn into a technology fanatic.
Not at all.
I actually think this opens a subtle window for ordinary people.
In the past, getting leverage was hard. You had to know how to code, be able to hire people, have a budget and have authority across departments.
That is no longer an absolute requirement.
Someone who thinks carefully about workflows can reorganize a small part of their work with ChatGPT, Claude, Codex, Zapier, Feishu tables or Notion automation, even without knowing how to write complex code.
At first, it will be clumsy.
And it will probably be slower than doing things by hand.
I mean it.
The first time you ask an agent to run a process, you may spend half a day adjusting the inputs, fixing errors and filling in context, only to find that doing it yourself would have taken five minutes.
That is when many people give up.
But there is a crucial difference.
Doing something by hand gets it done once.
Getting a workflow to run means doing less of it many times in the future.
That is a very plain lesson from the Industrial Revolution.
When steam engines first appeared, factories did not instantly become elegant and efficient. When electricity first entered factories, many simply replaced the old central drive shafts with electric motors. The layout stayed the same, so the efficiency gains were limited.
The real change came later, when people realized that if the source of power had changed, the way the factory was organized could change too.
I think the wave of AI agents is a little like that.
At first, people treated them as chat windows.
Write emails a little faster. Polish reports a little faster. Generate presentations a little faster.
That is useful, of course.
But it is only connecting a new motor to the old factory.
The more important next step is to rearrange the workflow.
Which tasks should a person initiate?
Which should an agent handle first?
Which judgments must remain a person’s responsibility?
Which quality standards can be written into the process?
Which lessons should not live only in the head of a longtime employee, but become a shared system for the team?
As for ClickUp, there is plenty to question about what it has done.
A 22% layoff is a real shock to the people who lost their jobs.
You cannot use an attractive story about the future to brush off their losses.
And a Gartner survey is a reminder that many companies using autonomous technologies did lay people off, but those layoffs did not necessarily bring meaningful financial returns.
That is awkward.
Some companies may just be using AI as shiny wrapping around the same old cost-cutting.
So I do not want to present ClickUp as a template for “the company of the future.”
There is no need.
I would rather treat it as a signal.
The signal is not necessarily a pleasant one.
But it is clear.
Companies are no longer satisfied with employees using ChatGPT.
They will start asking employees to prove that they can turn AI agents into dependable productivity.
Not a demo.
Not a flashy trick.
Not a message in the group chat saying, “I did it with AI in five minutes.”
It has to be reusable, verifiable and manageable.
Those three words are dull, but I think they matter a lot.
Reusable means it is not a one-off magic trick.
Verifiable means it is not merely impressive-looking when nobody knows whether it is right.
Manageable means it can become part of an organization rather than staying a clever personal hack.
That is one of my predictions for the next few years.
Many roles will not disappear right away.
But the way value is ranked inside them will change.
Take two people in operations. One can only execute an event plan. The other can make the whole process—from choosing the theme and preparing materials to scheduling, reviewing results and gathering user feedback—into a partly automated system.
Take two salespeople. One follows up with customers manually. The other uses agents to organize customer information, draft personalized follow-ups, flag risks and capture patterns behind successful deals.
Take two content creators. One struggles through a draft every day. The other builds a radar for useful source material, a topic pipeline, a fact-checking process and a headline-testing process.
The gap will keep growing.
And it is not about who is better at slacking off.
If anything, it will test people’s judgment more.
The stronger agents get, the less people can outsource their thinking.
You have to be clearer about what good looks like.
You have to be clearer about what must not be handed over.
You have to know which mistakes could be disastrous.
You have to be clearer about your taste, standards and limits of responsibility.
That is why I felt a little excited when I saw this news.
Not excited about layoffs.
That would be inhuman.
I am excited that organizations are finally beginning to imagine productivity differently.
For many years, talk of efficiency eventually turned into squeezing people: longer hours online, more meetings, denser OKRs, faster responses. Everyone was on a conveyor belt that never stopped.
But if agents can really take on some execution, could people get more time back for judgment, creativity, relationships, taste and responsibility?
Honestly, I do not know.
Maybe many companies will use this to squeeze people more cleverly.
Maybe many people will be forced to learn a completely unfamiliar way of working in very little time.
Maybe the million-dollar compensation band will belong to only a tiny handful of star employees, while most people face higher expectations and less security.
Those are all real concerns.
But if you ask me what an ordinary person should do now, I would say: do not panic.
And do not rely on a motivational rush, either.
A more realistic move is to start with one small task in your own work that is annoying, repetitive or easy to get wrong.
Do not try to redesign the whole department at once.
Pick one thing.
Maybe it is a weekly report.
Maybe it is the notes you write after every customer meeting.
Maybe it is the research you gather before writing an article.
Maybe it is the review spreadsheet you fill out after each campaign.
Break it down.
What goes in?
What judgments have to be made along the way?
What should come out?
Which parts must you review?
Which parts can ChatGPT or Claude handle first?
Then let it run once.
It is fine if it goes badly.
Fix it.
Write down what you had to fix.
Run it again next time.
This is clumsy, slow and not at all futuristic.
But a lot of things that really change people start this way.
Not with a beam of light shining down.
Just with you, on an ordinary afternoon, suddenly realizing: I do not have to copy and paste those 20 lines by hand anymore.
That is the moment the future really enters your workflow.
So I do not want to call ClickUp’s 22% layoffs simply good news or bad news.
They are more like a mirror.
They show how anxious companies are.
They also show how vulnerable individuals can be.
But the mirror reveals something else too.
The person who becomes more scarce in the future may not be the most obedient or the busiest.
It may be the person who can organize their judgment, experience and curiosity into a process that machines can take part in.
What a time to be alive, friends.
We used to talk about productivity as how fast one person could run.
In the future, it may mean how much nonhuman execution power one person can bring along toward a goal worth reaching.
That sounds a little cyberpunk.
But I really think it is getting close.
I write about AI, culture, going global and everything else that catches my interest. If you enjoyed this, please like, share and save. Thanks for reading. See you next time!